gpu: Do not relaunch the process during teardown.

content/browser/gpu/gpu_process_host.cc

Index: content/browser/gpu/gpu_process_host.cc
diff --git a/content/browser/gpu/gpu_process_host.cc b/content/browser/gpu/gpu_process_host.cc
index 16b1f0db35a5dbd3295720daeddaf209207af1e0..82875a87b2bfa051a02cae96e936221b22197129 100644
--- a/content/browser/gpu/gpu_process_host.cc
+++ b/content/browser/gpu/gpu_process_host.cc
@@ -369,6 +369,10 @@ GpuProcessHost* GpuProcessHost::Get(GpuProcessKind kind, bool force_create) {
   if (!force_create)
     return nullptr;
 
+  // Do not create a new process if browser is shutting down.
+  if (BrowserMainRunner::ExitedMainMessageLoop())
+    return nullptr;

[sadrul, 2017/04/10 15:18:00]

Note that this is running on the IO thread. The flag does get set [1] before the
thread-teardown starts [2]. But I am not entirely sure if this is OK to do in
all cases.

[1]
https://cs.chromium.org/chromium/src/content/browser/browser_main_runner.cc?t...
[2]
https://cs.chromium.org/chromium/src/content/browser/browser_main_runner.cc?t...

[piman, 2017/04/10 17:14:32]

Isn't this a data race, anyway?

[sadrul, 2017/04/10 17:51:50]

In the case when it's not terminating (i.e. in most general case), it is, yep.
(In the case when it is terminating, it's safe I think, since the main-thread is
blocked on the io thread shutting down.)

Would introducing a lock in BrowserMainRunner::ExitedMainMessageLoop() work?

+
   static int last_host_id = 0;
   int host_id;
   host_id = ++last_host_id;